faculty.georgetown.edufaculty.georgetown.edu/rtk8/9780199679935-kramer-revises1.pdf · comp. by:...

Comp. by: Jaganathan Stage : Revises1 ChapterID: 0002496108 Date:16/6/15Time:19:11:23 Filepath://ppdys1122/BgPr/OUP_CAP/IN/Process/0002496108.3dDictionary : OUP_UKdictionary 1

The Morphosyntax of Gender

OUP UNCORRECTED PROOF – REVISES, 16/6/2015, SPi


OXFORD STUDIES IN THEORETICAL LINGUISTICS

general editorsDavid Adger and Hagit Borer, Queen Mary, University of London

advisory editorsStephen Anderson, Yale University; Daniel Büring, University of California, LosAngeles; Nomi Erteschik-Shir, Ben-Gurion University; Donka Farkas, University ofCalifornia, Santa Cruz; Angelika Kratzer, University of Massachusetts, Amherst;Andrew Nevins, University College London; Christopher Potts, Stanford University;Barry Schein, University of Southern California; Peter Svenonius, University ofTromsø; Moira Yip, University College London

recent titles

43 Genericityedited by Alda Mari, Claire Beyssade, and Fabio Del Prete44 Strategies of Quantificationedited by Kook-Hee Gil, Steve Harlow, and George Tsoulas45 Nonverbal PredicationCopular Sentences at the Syntax–Semantics Interfaceby Isabelle Roy46 Diagnosing Syntaxedited by Lisa Lai-Shen Cheng and Norbert Corver47 Pseudogapping and Ellipsisby Kirsten Gengel48 Syntax and its Limitsedited by Raffaella Folli, Christina Sevdali, and Robert Truswell49 Phrase Structure and Argument StructureA Case Study of the Syntax–Semantics Interfaceby Terje Lohndal50 Edges in SyntaxScrambling and Cyclic Linearizationby Heejeong Ko51 The Syntax of Roots and the Roots of Syntaxedited by Artemis Alexiadou, Hagit Borer, and Florian Schäfer52 Causation in Grammatical Structuresedited by Bridget Copley and Fabienne Martin53 Continuations and Natural Languageby Chris Barker and Chung-chieh Shan54 The Semantics of Evaluativityby Jessica Rett55 External Arguments in Transitivity Alternationsby Artemis Alexiadou, Elena Anagnostopoulou, and Florian Schäfer56 Control and Restructuringby Thomas Grano57 The Interaction of Focus and Givenness in Italian Clause Structureby Vieri Samek-Lodovici58 The Morphosyntax of Genderby Ruth KramerFor a complete list of titles published and in preparation for the series, see pp XXX–X.


hodgecHighlight


The Morphosyntaxof Gender

RUTH KRAMER

1



3Great Clarendon Street, Oxford ox2 6dp

United Kingdom

Oxford University Press is a department of the University of Oxford.It furthers the University’s objective of excellence in research, scholarship,

and education by publishing worldwide. Oxford is a registered trade mark ofOxford University Press in the UK and in certain other countries

# Ruth Kramer 2015

The moral rights of the author have been asserted

First Edition published in 2015Impression: 1

All rights reserved. No part of this publication may be reproduced, stored ina retrieval system, or transmitted, in any form or by any means, without the

prior permission in writing of Oxford University Press, or as expressly permittedby law, by licence, or under terms agreed with the appropriate reprographics

rights organization. Enquiries concerning reproduction outside the scope of theabove should be sent to the Rights Department, Oxford University Press, at the

address above

You must not circulate this work in any other formand you must impose this same condition on any acquirer

Published in the United States of America by Oxford University Press198 Madison Avenue, New York, NY 10016, United States of America

British Library Cataloguing in Publication DataData available

Library of Congress Control Number: 2014958986

ISBN 978–0–19–967993–5 (Hbk)ISBN 978–0–19–967994–2 (Pbk)

Printed and bound byCPI Group (UK) Ltd, Croydon cr0 4yy

Links to third party websites are provided by Oxford in good faith andfor information only. Oxford disclaims any responsibility for the materials

contained in any third party website referenced in this work.



Contents

General preface ixAcknowledgments xList of figures and tables xiiList of abbreviations xiii

1 Introduction 1

1.1 Major themes 11.2 Frameworks: Minimalism and Distributed Morphology 3

1.2.1 Setting the scene 31.2.2 A little Minimalism, a lot of DM 41.2.3 Lexical decomposition 71.2.4 Summary 9

1.3 Limitations 101.4 The lay of the land: chapter previews 11

2 The Amharic gender system and previous approaches to gender 14

2.1 Introduction 142.2 Gender in Amharic 14

2.2.1 The facts 152.2.2 Summary, typology, diachrony 20

2.3 The morphosyntax of gender: previous approaches 232.3.1 GenP and NumP 242.3.2 Gender on the noun: lexicalism 262.3.3 Gender on the noun: Distributed Morphology 32

3 A n analysis of gender 37

3.1 Introduction 373.2 The morphosyntax of gender: a new approach 373.3 Further evidence for n as the locus of gender 423.4 Licensing conditions and interpretability: refining the analysis 49

3.4.1 Licensing conditions 503.4.2 The interpretability of gender features 57

3.5 Conclusion 63



4 Defining gender 65

4.1 What gender is, and what gender isn’t 654.1.1 Classifiers and declension class are not genders 674.1.2 Noun class and pronominal gender are gender 684.1.3 Gender is not like other phi-features 69

4.2 Conclusion 71

5 Case study 1: Two genders, three ns 72

5.1 Introduction 725.2 Masculine default or feminine default 73

5.2.1 Masculine default: Dieri 745.2.2 Feminine default: Zayse and Zargulla 79

5.3 Animacy-based gender systems and a negative prediction 825.3.1 Introduction 825.3.2 The gender system of Lealao Chinantec: description 845.3.3 The gender system of Lealao Chinantec: analysis 86

5.4 Conclusion 88

6 Case study 2: Adding an uninterpretable gender feature 89

6.1 Introduction 896.2 Uninterpretable feminine gender: Spanish 90

6.2.1 The gender system of Spanish: description 906.2.2 The gender system of Spanish: analysis 956.2.3 Comparing Spanish and Amharic 98

6.3 Uninterpretable masculine gender: Maa 996.3.1 The gender system of Maa: description 1006.3.2 The gender system of Maa: analysis 104

6.4 Uninterpretable animacy: Algonquian 1056.4.1 The gender system of Algonquian: description 1056.4.2 The gender system of Algonquian: analysis 110

6.5 Conclusion 114

7 Case study 3: Three-gender languages 116

7.1 Introduction 1167.2 Three genders, three ns: Mangarayi 116

7.2.1 The gender system of Mangarayi: description 1177.2.2 The gender system of Mangarayi: analysis 118

7.3 Three genders, one uninterpretable feature: Wari’ 1217.3.1 The gender system of Wari’: description 1237.3.2 The gender system of Wari’: analysis 125


vi Contents


7.4 Three genders, two uninterpretable features: Lavukaleve 1297.4.1 The gender system of Lavukaleve: description 1317.4.2 The gender system of Lavukaleve: analysis 1337.4.3 Conclusion: Lavukaleve 138

7.5 Conclusion 1397.6 Excursus: default gender in three-gender languages 140

7.6.1 Default gender in Mangarayi, Wari’, and Lavukaleve 1417.6.2 Default gender in Russian, Icelandic, and Tamil 1427.6.3 Suggestions for analysis and conclusion 145

8 Gender is not on Num: Evidence from Somali and Romanian 147

8.1 Introduction 1478.2 Gender switch in Somali: all plurals are n 148

8.2.1 Gender in Somali: description and analysis 1498.2.2 The plural system of Somali 1518.2.3 Plurality is on n in Somali: evidence 1558.2.4 Somali plural system: analysis 1598.2.5 Alternative analyses and conclusions 164

8.3 Gender switch in Romanian: the neuter is real 1668.3.1 The Romanian gender system 1668.3.2 The analysis of Romanian gender 1698.3.3 Alternative analyses 1778.3.4 Interim conclusion 183

8.4 Conclusion 183

9 Gender and nominalizations 185

9.1 Introduction 1859.2 Nominalizations are gendered: data 185

9.2.1 Theoretical background 1859.2.2 Various types of nominalizations are gendered 1879.2.3 All of the gender-relevant ns can nominalize 190

9.3 Nominalizations are gendered: predictions 1929.3.1 Derived nouns without n 1939.3.2 One gender, many exponents 1969.3.3 Nominalizations in an animacy gender system 2009.3.4 Interim summary 204

9.4 Two problems (and their solutions) 2059.4.1 Gender features exponed separately? 2059.4.2 Same nominalization, different genders? 209

9.5 Conclusion 212


Contents vii


10 The highest gender wins and the interaction of gender anddeclension class 213

10.1 Introduction 21310.2 May the highest gender win: gender and multiple ns 213

10.2.1 The highest gender wins: data 21410.2.2 A cyclicity explanation for the highest gender hypothesis 22010.2.3 Conclusion and implications 226

10.3 A diminutive digression 22710.4 Declension class and gender 233

10.4.1 Background 23310.4.2 Spanish declension class: a case study 23510.4.3 Summary, and directions of correlations 243

10.5 Conclusion 244

11 Conclusion 245

11.1 Putting it all together 24511.1.1 Question A: Gender on n 24511.1.2 Question B: The relationship between natural and arbitrary

gender 24711.1.3 Question C: Gender assignment 24811.1.4 Evidence for Distributed Morphology 250

11.2 Areas of future research 25011.2.1 Phonologically determined gender assignment 25011.2.2 Languages with more than three genders 25111.2.3 The role of social factors 25311.2.4 Languages that lack gender 255

11.3 Meta-conclusion 256

References 257Language Index 281Subject Index 284


viii Contents

hodgecCross-Out

hodgecReplacement Text4

hodgecInserted Text and language family


General preface

The theoretical focus of this series is on the interfaces between subcomponents of thehuman grammatical system and the closely related area of the interfaces between thedifferent subdisciplines of linguistics. The notion of “interface” has become central ingrammatical theory (for instance, in Chomsky’s Minimalist Program) and in linguis-tic practice: work on the interfaces between syntax and semantics, syntax andmorphology, phonology and phonetics, etc. has led to a deeper understanding ofparticular linguistic phenomena and of the architecture of the linguistic componentof the mind/brain.

The series covers interfaces between core components of grammar, includingsyntax/morphology, syntax/semantics, syntax/phonology, syntax/pragmatics,morphology/phonology, phonology/phonetics, phonetics/speech processing, seman-tics/pragmatics, and intonation/discourse structure, as well as issues in the way that thesystems of grammar involving these interface areas are acquired and deployed in use(including language acquisition, language dysfunction, and language processing). Itdemonstrates, we hope, that proper understandings of particular linguistic phenom-ena, languages, language groups, or inter-language variations all require reference tointerfaces.

The series is open to work by linguists of all theoretical persuasions and schools ofthought. A main requirement is that authors should write so as to be understood bycolleagues in related subfields of linguistics and by scholars in cognate disciplines.

Gender features stand at the intersection of syntax, semantics, and morphology,and how their role is divided between these three domains has long been a puzzle. Inthis monograph, Ruth Kramer develops a new theory of the morphosyntax of gender,arguing that gender features appear not on lexical roots or on elements high in thestructure of noun phrases, but on the nominalizing head n. She defends a particulartheory of gender features based on the notions of interpretability and bivalence, andshows how this theory predicts the attested typological variation in gender systems,drawing on fascinating data from a number of lesser studied languages. Overall, thebook proposes the first comprehensive theory of gender as a phi-feature in generativelinguistics.

David AdgerHagit Borer



Acknowledgments

This project has been through various stages of development for many years, and it isa challenging task to acknowledge all of the generous people who have assisted alongthe way. Nevertheless, I will do my best since they all greatly deserve it.

The core of this book originates in my dissertation (Kramer 2009), and I can neverthank my committee enough for their help and insights: Sandra Chung, JorgeHankamer, and James McCloskey. I also thank my peers at Santa Cruz for theirfeedback and encouragement (especially the members of the Morphology ReadingGroup): Vera Gribanova, Boris Harizanov, Mark Norris, Kyle Rawlins, Anie Thomp-son, and Matthew Tucker.

I have presented portions of this book at the University of Pennsylvania, YaleUniversity, the ST@R reading group at Rutgers University, the syntax–semanticsreading group at New York University, the University of Delaware, the University ofCalifornia, Santa Cruz, and George Mason University. I have also presented portionsof it at the following conferences/workshops: the 15th International MorphologyMeeting, the 30th West Coast Conference on Formal Linguistics, the 43rd and 45thAnnual Conferences on African Linguistics, and the workshop on “Allomorphy: ItsLogic and Limitations.” I thank the audiences at all of these presentations for theirinvaluable feedback and suggestions, especially Pranav Anand, Karlos Arregi, MarkBaker, Ricardo Bermúdez-Otero, Amy Rose Deal, Marijke de Belder, Rose-MarieDéchaine, David Embick, Jane Grimshaw, Richard Kayne, Jean Lowenstamm,Andrew Nevins, Elizabeth Ritter, Peter Svenonius, Martina Wiltschko, Matt Wolf,and Raffaella Zanuttini.

Portions of Chapters 2 and 3 appear in “Gender in Amharic: a morphosyntacticapproach to natural and grammatical gender,” Language Sciences 43: 102–15. Portionsof Chapters 3 and 8 are to appear in “A split analysis of plurality: evidence fromAmharic,” to be published in Linguistic Inquiry. I thank the four anonymousreviewers of these papers for comments and suggestions that have been integratedinto this book as well. I also thank the members of the Seminar on Morphosyntax atGeorgetown in Spring 2014 for reading “Gender in Amharic” and offering usefulfeedback: Melanie Ashkar, Colleen Diamond, Ava Irani, Morgan Rood, Brett Sutton,and Katherine Vadella.

I turn next to those who engaged with the book as a manuscript. First credit goes toMark Norris, for reading almost the entire manuscript with incredible speed andcheer, and for assisting with Icelandic and Estonian data. I am also deeply gratefulto the following readers/commenters on portions of the manuscript: Azeb Amha,Peter K. Austin, Joshua Birchall, Héctor Campos, Sandra Chung, Donka Farkas,



Christopher Green, Vera Gribanova, Boris Harizanov, Heidi Harley, Bryce Huebner,Donna Lardiere, Eric Mathieu, Jason Merchant, Francesca Merlan, Doris Payne, PaulPortner, Conor Quinn, James Rupp, Angela Terrill, and Katherine Vadella. None ofthese people are responsible for errors of fact or interpretation.

I am also profoundly thankful to all the consultants whose insight has shaped thiswork: Senayit Ghebrehiywet, Girma Demeke, Bekale Seyum, Betselot Teklu, HileenaEshetu, Harya Tarakegn, Mehret Getachew Tadesse, Meriem Tikue, Azeb Amha,Marta Baffy, Héctor Campos, Angela Donate Velasco, Donka Farkas, Paula Ganga,Nagarajan Selvanathan, and Esther Surenthiraraj.

Additionally, I am grateful for the research support that I have received during thecreation of this book. I thank Georgetown University for a Junior Faculty ResearchFellowship and Summer Academic Grant that supported several months of writing,and I also thank Carole Sargent and the book club members of Summer 2014 forhelping me keep up the momentum. Credit is also due for research assistance toLindley Winchester, Morgan Rood, Laura Siebecker, and Lauren McGarry.

Finally, this book could never have been completed without the patient support ofmy family and of Bryce Huebner. This book is dedicated to them.


Acknowledgments xi


List of figures and tables

Figures

1.1 The DM model of the grammar 5

1.2 The architecture of PF 6

8.1 Controller genders in Romanian 178

Tables

3.1 How roots are licensed in Amharic 50

3.2 Amharic licensing conditions 57

5.1 Two genders, three ns 72

5.2 Definite determiners/3rd person singular pronouns in Dieri 75

5.3 Predicted genders 87

6.1 Two genders, four ns 89

6.2 Licensing conditions for Spanish 97

6.3 Licensing conditions for Maa 105

7.1 Four ns: possible inventories 121

7.1 Four ns: possible inventories (with genders) 122

7.3 Lavukaleve object marker paradigm 136

8.1 Plural strategies in Somali 154

8.2 Analysis of the Somali plural 162

10.1 Analysis of the Somali plural (identical to 8.2) 224

10.2 Spanish declension classes 236

11.1 Possible inventories of gender-relevant ns 248

11.2 Noun classes in Sesotho 252


hodgecCross-Out

hodgecReplacement Text2


List of abbreviations

1 1st person

2 2nd person

3 3rd person

abs absolutive

acc accusative case

act action particle

adj adjectivalizer

affirm affirmative

an(im) animate

ana anaphoric

ant anterior verb suffix

aux auxiliary

caus causative

cl classifier

coll collective

cop copula

d determiner

dat dative

def definite marker

dim(in) diminutive

dis distant demonstrative

dm Distributed Morphology

dur durative

emph emphatic particle

erg ergative case

f feminine

f/n feminine or neuter

fem female natural gender

foc focus

gc gender clitic

gen genitive case



hab habitual

hon honorific

imp imperative

inan inanimate

inf infinitive

infl inflection

int intensive

loc locative case or locative marker

m masculine

m/f masculine or feminine

m/n masculine or neuter

n neuter

n ‘little’ n, nominalizing head

neg negation

nf non-feminine

nmlz nominalizer

nom nominative case

nonfin non-finite

.o object agreement/marker

past past tense

pc past continuous

pl plural

poss possessive

poss.art possessive article

pp past punctual

prep preposition

pres present tense

prod product verbalizer

prop proprietive case

red reduplicant

refl reflexive

rf realis future

rp/p realis past/present

s singular

.s subject agreement/marker


xiv List of abbreviations


sta stative

sup supine

t tense

top topic

tv theme vowel

vblz verbalizer

vicin vicinity


List of abbreviations xv


1

Introduction

1.1 Major themes

Gender is regularly defined as the sorting of nouns into two or more classes, asreflected in agreement morphology on determiners, adjectives, verbs and othersyntactic categories (e.g. Hockett 1958: 231, Fodor 1959: 2, Corbett 1991: 1, Comrie1999: 457, Matasović 2004: 19–20). Consider the Amharic examples in (1)ab.

(1) a. ya säw dägg näw b. yatʃtʃ set dägg natthat.m man good be.3ms.s that.f woman good be.3fs.s‘That man is good.’ ‘That woman is good.’ (Leslau 1995:66,67)

In (1)a, the demonstrative is ya and the copular verb is näw, whereas in (1)b, thedemonstrative is yatʃtʃ and the copular verb is nat. Since the demonstrative and thecopular verb formally differ depending only on the head noun of the subject, it isclear that säw ‘man’ belongs to one gender (masculine) and set ‘woman’ belongs toanother gender (feminine).1

Gender has been called “a time-honored subject of linguistics” (Unterbeck andRissanen 2000: ix), and Matasović (2004: 13) dubs it “the only grammatical categorythat ever evoked passion.” Corbett calls gender the “most puzzling of the grammat-ical categories” (Corbett 1991: 1). Unsurprisingly then, there are rich and significantliteratures on the sociolinguistics of gender (see e.g. Hellinger and Bußmann 2001),the acquisition and processing of gender (see e.g. Franceschina 2005), the typology ofgender systems (see e.g. Corbett 1991), and the diachronic development and loss ofgender systems (see e.g. Matasović 2004).

However, there has been less research on the morphological and syntactic aspectsof gender. There are clear, thorough descriptions of gender assignment inmany languages (see e.g. Corbett 1991: chs 2 and 3), but the most basic questionsfor a morphosyntactic analysis of gender assignment remain controversial: whereis gender located in the hierarchical structure? How is gender assignment

1 See Chapter 4 for a refinement of this definition of gender.

The Morphosyntax of Gender. First Edition. Ruth Kramer.© Ruth Kramer 2015. Published 2015 by Oxford University Press.



morphosyntactically accomplished? Is gender lexically listed on a noun, with rela-tively little role for the syntax to play? Or is there a gender projection in the syntaxthat combines with nouns to assign gender? (And if so, what is that projection?)

Additionally, the morphosyntactic literature on gender tends to downplay therelationship between natural gender (gender based on some semantic property, e.g.male/female, animate/inanimate) and (what I will call) arbitrary gender, that is,gender assigned without reference to any semantic property. The arbitrary genderassigned to a noun often varies across languages. For example, the noun ‘morning’has a different gender in French (masculine), Hausa (feminine), and Russian(neuter).

(2) a. French b. Hausa c. Russianmatin sāfiyā utromorning.m morning.f morning.n‘morning’ ‘morning’ (Newman 2000: 204) ‘morning’

There is nothing about the meaning of ‘morning’ that requires it to have a particulargender in any of these languages; its gender is assigned arbitrarily. However, there areother nouns whose meaning does determine their gender in many languages. Forexample, the word for ‘father’ is masculine in all three languages, because nouns thatrefer to male entities are (generally) masculine in all of these languages.

(3) a. French b. Hausa c. Russianpère ùbā otecfather.m father.m father.m‘father’ ‘father’ (Newman 2000: 201) ‘father’ (Corbett 1991: 34)

Gender assignment therefore operates in two dimensions: gender is assigned accord-ing to some natural/semantic property of the real world, or it is assigned arbitrarily.In many languages, gender is assigned only according to some natural property of thenoun (e.g. Dieri (Chapter 5), Mangarayi (Chapter 7), Tamil (Arden 1942, Asher1985)).

Nevertheless, not all morphosyntactic approaches to gender treat natural gender ascentral to the analysis; many either set natural gender aside, or “convert” it via a ruleinto the same type of feature as arbitrary gender. This may be because the morpho-syntactic relationship between arbitrary gender and natural gender has been exploredfor only a few related languages which all have gender systems heavily based onarbitrary gender (e.g. Spanish, Italian, Greek). It remains unclear how both types ofgender are expressed via the same morphological resources, whether they have thesame syntactic location, and even whether they use the same set of features.

This book aims to fill these gaps in the morphosyntactic literature. I develop atheory of gender assignment that specifies the locus of gender features in the syntaxas the nominalizing head n. A handful of previous analyses have proposed that


2 1 Introduction


gender is on n (see e.g. Kihm 2005, Ferrari 2005, Lecarme 2002, Lowenstamm 2008,Acquaviva 2009, Kramer 2009), but I significantly expand on these approaches.I adduce evidence for a n locus from a detailed case study of Amharic, whichfurnishes unique evidence for a n locus because of its unusual plural system.I argue that gender features are never on Num by studying two cases of genderswitch in the plural (Somali and Romanian), and offer further evidence that gender ison n by examining the relationship between gender and nominalization, and genderand declension class, cross-linguistically.

None of the previous n-analyses systematically address the question of naturalversus arbitrary gender, so in this book, I connect the n-based analysis explicitly tothe two different types of gender: natural gender is an interpretable feature on n,whereas arbitrary gender is an uninterpretable version of the same feature on n.I develop a theory of gender features on n, focusing on how the gender features in thesyntax map onto the gender features relevant for morphological exponence. Buildingon this approach to gender features, I make positive and negative predictions aboutpossible gender systems, and show that they are borne out. Overall, the bookprovides, within the Chomskyan tradition, the first large-scale, cross-linguisticallyoriented analysis of the morphosyntax of natural and arbitrary gender.

In the remainder of the Introduction, I first provide some background on theframeworks adopted in the book (Section 1.2). I discuss several topics that the bookwill not address (and why not) in Section 1.3, and then preview the main claims of thebook and the contents of each chapter in Section 1.4.

1.2 Frameworks: Minimalism and Distributed Morphology

1.2.1 Setting the scene

Because this book is an investigation of the morphosyntax of gender, it is necessary toidentify the frameworks adopted for syntax and morphology. I take a theoreticalapproach that is generative and broadly Chomskyan, adopting Minimalist syntax andDistributed Morphology. There are numerous influential approaches to gender bothoutside of the generative tradition (see e.g. Lakoff 1987) and in non-Chomskyanframeworks (Network Morphology: Corbett and Fraser 2000a, 2000b, Evans et al.2002; HPSG: Wechsler and Zlatić 2003; Optimality Theory: Rice 2006; and manyothers). Nevertheless, I take a generative Chomskyan approach for several reasons.

First of all, there is not a universally accepted morphosyntax for gender within theChomskyan tradition. Consequently, the book aims to develop a benchmark analysisfor gender, and to unite (or find reasons to set aside) previous Chomskyan analyses.Since a Chomskyan approach is the mainstream within generative linguistics, thelack of a standard analysis of gender is striking. Moreover, there has been less cross-linguistic work done within the Chomskyan tradition on gender. The main languages


1.2 Frameworks 3


investigated have been from the Indo-European and Bantu families, but little atten-tion has been paid to languages like Amharic (Chapters 2 and 3), Dieri (Chapter 5),Wari’ (Chapter 7), and Lavukaleve (Chapter 7). This book aims to bring a larger spanof data to inform a Chomskyan approach to gender.

Additionally, one of the most fundamental questions in a generative syntacticframework is the relationship between the lexicon and the syntax. Is there a verypowerful lexicon containing word-related idiosyncratic information as well as havinga generative capacity (lexicalism), or are the capabilities and information associatedwith the lexicon distributed throughout the grammar (Distributed Morphology)?Gender assignment seems like a quintessentially “lexical” phenomenon, since it canbe very arbitrary. However, the analysis developed in this book provides a way toassign gender without a traditional lexicon. As a result, the analysis contributes to thedevelopment of linguistic theory in that it provides support for a non-lexicalapproach to morphology.

A final, more practical motivation for using Minimalism and Distributed Morph-ology is that they are mutually consistent. Their assumptions do not conflict, andthey are capable of being used in tandem without much tension. This is notnecessarily true for other theories of morphology, especially lexicalist theories sincelater Minimalism has moved away from lexicalism (Chomsky 2000, 2001, 2004).

Overall, I do not use this book to argue that a Chomskyan approach to gender isbest among all other approaches—it is merely the framework that I assume for theinvestigation. However, hopefully, the explicit proposals here will serve as a goodstarting point for cross-framework comparison. In the remainder of this section,I present the details of my theoretical assumptions.

1.2.2 A little Minimalism, a lot of DM

On the syntactic side, I assume the conventional Principles and Parametersapproach to syntax, adopting the Minimalism of Chomsky (2000, 2001, 2004) inparticular. I rely on the phase as a cyclic unit in Chapter 10, and it is largelyexplained within that chapter. I also occasionally mention the syntactic relationAgree, i.e. a relation between a probe with unvalued features and a goal in its c-command domain with valued features, whereby the probe’s features are valued bythe goal (Chomsky 2001, Pesetsky and Torrego 2007). However, I do not assumethat DP-internal agreement (also called “concord”) is necessarily accomplished viaAgree (see e.g. Norris 2014 for a non-Agree approach to concord). Accordingly,I often refer to any agreement relation informally as just agreement, not the Agreerelation in particular.

In general, I use theory-neutral terminology from Corbett (2006a) to describe theparticipants in an agreement relationship. The agreement controller is the elementthat determines the agreement; it is born into the derivation already containing the


4 1 Introduction


relevant features. The agreement target is the element whose form is determined byagreement; its features are valued or altered through agreement. For example, in (1)b,set ‘woman’ is the agreement controller, and the demonstrative and the copular verbsare agreement targets.

Because it is less well known than Minimalism, and more prominent in theargumentation of the book, I will spend the majority of this section outliningDistributed Morphology (henceforth DM; Halle and Marantz 1993). DM started tobe developed as a theory of morphology in the early 1990s. Seminal works includeHalle (1990), Halle and Marantz (1993, 1994), Harley and Noyer (1999), Embick andNoyer (2001), Embick and Noyer (2007), and Harley (2014), and many others.

Distributed Morphology maintains that there is no centralized lexicon. Instead,word formation occurs either in the syntax via head movement or at PF viamorphology-specific operations. The information contained in the lexicon in othertheories (phonological information, semantic information, category, and syntacticfeatures) is ‘distributed’ throughout the grammar in various lists. Figure 1.1 showswhen these lists are accessed, and it also illustrates the DM model of the grammar.The lists are in circles.

Starting at the top, there is a list of syntactic terminal nodes which are manipulatedby operations during the syntactic derivation. The terminal nodes are bundles ofmorphosyntactic features (including category), and these bundles lack any kind ofmorphophonology in the syntax (this ensures “Phonology-Free Syntax”; cf. Zwickyand Pullum 1986). The feature bundles are often called “morphemes” in DM,and throughout the book, I use the terms “feature bundle” and “morpheme”interchangeably.

After the syntactic derivation is complete, the derivation is sent to PhonologicalForm and Logical Form. Logical Form operates according to standard formal seman-tic assumptions (compositionally, with operations like quantifier raising etc.). TheEncyclopedia is another one of the DM lists that houses information that used to bein the lexicon—in the case of the Encyclopedia, it is the semantic information. AsHarley (2014) phrases it, the Encyclopedia provides “instructions for interpreting

Syntactic terminal nodes

Vocabulary Items Syntactic derivation (narrow syntax) Encyclopedia

Morphology

LFPF

Figure 1.1 The DMModel of the grammar (inspired by Harley and Noyer 1999: 3, Embick andNoyer 2007: 22, Kelly 2013: 12)


1.2 Frameworks 5


terminal nodes in context” (p. 228). For example, the Encyclopedia contains theinstructions to interpret the root √cat in a nominal context as a “feline mammal,four legs, meows, subject of many internet memes, etc.”

Turning now to the PF branch, “Morphology” is the subcomponent of thegrammar where morphological operations occur. One of these operations is Vocabu-lary Insertion, whereby syntactic feature bundles are given morphophonologicalcontent, i.e. they are exponed by the insertion of Vocabulary Items (this idea isreferred to as Late Insertion; cf. the Separation Hypothesis of Beard 1995a).

There are several stages of PF, as shown in Figure 1.2. Directly after the PF/LFBranching, hierarchical structure is still present and certain morphological oper-ations that manipulate feature bundles can occur. Such operations include Lowering,which lowers one feature bundle to adjoin to another (Embick and Noyer 2001),Fission, which splits off a feature from a feature bundle and grants it its own node(Noyer 1997 among many others), and Fusion, which combines two feature bundlesinto one node (Halle 1997 among many others).

Next, the Vocabulary Items are inserted (the terminal nodes are provided withphonological content), and the structure is linearized. Various post-linearizationoperations (e.g. Local Dislocation, a switch in linear order between two nodes;Embick and Noyer 2001, Embick 2003) also take place, and these operations areconditioned by precedence relations. Finally, prosodic domains are built, and thePF derivation finishes with a complete phonological and linear representation(although see Pak 2008 on how the prosodic part of the model may presentdifficulties).

Vocabulary Insertion is the process whereby it is decided which Vocabulary Itemshould be inserted at a particular feature bundle. It will play a large role in the book,so it is worthwhile to spend some further time on it. Vocabulary Insertion is a verylocal process: it applies to one feature bundle at a time, and only one Vocabulary Itemcan be inserted for any given feature bundle. A Vocabulary Item is a relation between

(Syntactic derivation)⇓

PF/LF Branching

Lowering, Fission, Fusion, etc. ← Hierarchical arrangement of morphemes

Vocabulary Insertion ← Linearization imposed by Vocabulary Insertion

Building of prosodic domains

Phonological Form

Figure 1.2 The architecture of PF (inspired by Embick and Noyer 2001: fig. 1)


6 1 Introduction


a phonological string and information about where the string can be inserted. Theinformation about where the string is inserted is made up both of features andcontextual restrictions. Some Vocabulary Items for the past tense in English areshown in (4).

(4) a. T, [past] $ -t / { √leave, √bend, . . . }b. T, [past] $ -ed (Embick and Marantz 2008)

The Vocabulary Items in (4) are in competition to realize the feature bundleT, [past]. Two main principles determine which Vocabulary Item wins a givencompetition: the Pān ̣inian Principle (also known as the Elsewhere Condition) andthe Subset Principle (Halle 1997). The Pāṇinian Principle states that a more specificrule is applied before a less specific rule, and it suffices to determine the winner in (4).If the context is met to insert Vocabulary Item (4)a (i.e. the root is √leave or √bend),then it must be inserted, since it is more specific than Vocabulary Item (4)b (becauseit has a contextual restriction). The Subset Principle determines the winner in othercases.

(5) Subset Principle

(i) The phonological exponent of a Vocabulary Item is inserted into a positionif the item matches all or a subset of the features specified in that position.

(ii) Insertion does not take place if the Vocabulary Item contains features notpresent in the morpheme.

(iii) Where several Vocabulary Items meet the condition for insertion, the itemmatching the greatest number of features specified in the terminal mor-pheme must be chosen. (Halle 1997: 428)

The Subset Principle ensures that a Vocabulary Item cannot be inserted thatcontains features not present in the morpheme. However, the Vocabulary Itemmight contain fewer features than are present in the morpheme, in which case theVocabulary Item is referred to as “underspecified.” The Subset Principle also statesoutright that Vocabulary Items that match the most features with the givenmorpheme win.

Many of the concepts just introduced will be discussed in more detail at variouspoints throughout the book. However, I hope that this sketch of the core ideassuffices before they are presented in context.

1.2.3 Lexical decomposition

The assumptions in Section 1.2.2 form the core of DM, but much DM research hasalso pursued the idea that lexical categories are composed of a category-neutralroot and a category-determining head, an idea also known as lexical decomposition(see e.g. Marantz 1997, 2001, Arad 2003, 2005, Embick and Noyer 2007, Embick and


1.2 Frameworks 7


Marantz 2008, Harley 2014).2 For example, a verb like hammer consists of a root√hammer, and a functional head v that “verbalizes” it.

(6) vP Category Neutrality: Verbal

v √P

√hammer

This also goes for nouns, which consist of a root and the nominalizing functionalhead n.

(7) nP Category Neutrality: Nominal

n √P

√hammer

(7) results in the nominal hammer ‘a tool for pounding nails’ whereas (6) results inthe verb hammer ‘to pound (something)’.

The empirical motivation for lexical decomposition is provided by the well-knowndifferences between lexical and syntactic word formation. Word formation in thelexicon is more prone to phonological and semantic irregularities (e.g. specialphonological processes, idiomatic meanings), whereas syntactic word formation ismorphophonologically regular and has semantically predictable meaning. However,in DM, there is no lexicon or lexical processes, so the contrast must be captured in adifferent way. Marantz (2001, 2007) and Arad (2003, 2005) specifically propose that“lexical” word formation corresponds to word formation from roots—the combin-ation of a category-defining head (n, v) with a root. “Syntactic” word formationcorresponds to word formation from words—the combination of some head with acategorized word (i.e. nP, vP). In subsequent research, the predictions and conse-quences of this proposal have been explored (see e.g. Embick 2010 on contextualallomorphy). Even though some of the original claims have been disputed (see e.g.Harley 2014 on the domain of semantic interpretation), lexical decomposition hasbecome a major approach to syntactic categories in modern morphosyntax. The aimof this book is to explore the role of the categorizing head n in expressing genderfeatures in nominals.

Throughout the book, the roots in syntactic trees are labeled by a word in thelanguage in question after a square-root sign, as in (6) and (7). In early DM, however,

2 It must be noted that this idea was not entirely new (see e.g. van Riemsdijk 1990 on n), nor do onlyDistributed Morphologists subscribe to it (see e.g. Lowenstamm 2008). See also Borer (2005) for a similarapproach, although she argues (pp. 20–21) against the specific DM analysis adopted here.


8 1 Introduction


roots were indicated only by a square-root sign in the syntax, and were assumed to beundifferentiated. Marantz (1995) argued that roots do not compete for VocabularyInsertion, i.e. that any root could be inserted at a root node (as long as its licensingconditions were met). In other words, to the grammar, it does not matter whether‘hammer’ or ‘wrench’ is inserted at a root node, just as long as both of them arelicensed under a n.

This approach to roots, however, makes the prediction that there will be no rootsuppletion. If roots do not compete for insertion, then different forms of a root willnot compete with each other for insertion at a particular context (e.g. the VocabularyItem person will not be able to compete with people in a plural context). Although theidea that roots never undergo suppletion has been defended (see e.g. Marantz 1997),I agree with those who have argued that root suppletion is attested (see especiallyHarley 2014), and therefore I do not subscribe to the idea that roots do not compete.

Nevertheless, the representations in (6) and (7) suggest that the phonologicalcontent of a root is present from the beginning of the derivation. If this is true,then there is again a problem with root suppletion (if the phonological content of aroot is present from the beginning of a derivation, how could roots ever be supple-tive?—see Harley 2014). A solution is at hand, though: I assume that roots areidentified by non-phonological indices/labels in the syntax, following much recentresearch (Pfau 2000, 2009, Embick 2000, Embick and Noyer 2007, Acquaviva 2009,Harley 2014). The index for a root is exponed at PF via Vocabulary Insertion (wheredifferent root allomorphs can compete for insertion) and interpreted at the Encyclo-pedia in a given context. I will represent indices numerically. (8) shows the syntacticrepresentation, Vocabulary Item, and one possible interpretation of the root identi-fied as √hammer in (6) and (7).

(8) a. √169 syntactic representationb. √169 $ hammer Vocabulary Itemc. [nP n [ √169]] is interpreted as a type of tool with a Semantic interpretationlong narrow handle and a hard, specially shaped head,often used for pounding nails, etc . . .

I will continue to represent roots in individual trees and in the prose with the wordsof the language in question, but this is only for expository/clarificational purposes.

1.2.4 Summary

Overall, I assume the syntax operates as per Minimalist assumptions, and I take aDistributed Morphology approach to the grammar. This includes distributing listedinformation across the grammar, and having syntactic feature bundles be exponedpost-syntactically via Vocabulary Insertion. I also assume lexical decomposition: lexicalcategories are decomposed into category-neutral roots and category-defining heads.


1.2 Frameworks 9


1.3 Limitations

This book aims to develop a broadly applicable analysis of gender, but there arecertain necessary limitations on its scope. For example, the book does not engagedirectly with most of the puzzles related to gender agreement (see e.g. Wechsler andZlatić 2003, Carstens 2011, Matushansky 2013). From one perspective, gender agree-ment is a separate phenomenon from gender assignment. The mechanisms thatenforce gender agreement have been the focus of much morphosyntactic researchon their own, i.e. research that is not concerned with gender assignment. However,from another perspective gender agreement is closely related to gender assignment.Most nominals do not have overt gender markers, so the only evidence that a nounhas a particular gender is the gender agreement with that noun on other categories.The discussion of gender agreement in this book will therefore be limited to puzzlesthat relate to gender assignment.

Overall, the book will not advocate a particular theory of gender agreement. Therewill be little discussion of the mechanics of agreement (whether it requires a syntacticAgree relation or is entirely post-syntactic, whether the mechanisms for verbalagreement and DP-internal agreement are different, etc.). I also largely set asidepuzzles relating to hybrid nouns (Corbett 1991: ch. 8, Matushansky 2013) and togender resolution (Corbett 1991: ch. 9), i.e. how conjoined noun phrases agree wheneach noun phrase has a different gender. The most sustained discussion of genderagreement is in Chapter 10, where it is argued that agreement is with the highestgender feature in the nominal phrase.

The book also does not engage with gender-related puzzles of nominal ellipsis (seee.g. Bobaljik and Zocca 2011, Merchant 2014). The general approach to gender inBobaljik and Zocca (2010) is very compatible with the analysis of gender developed inthe book, and at the end of Chapter 8, I briefly discuss how some results fromnominal ellipsis support the idea that gender is not on Num. However, fleshing outthe details requires more focus on ellipsis than there is space for here, and I leave thistopic for future research.

The book also is limited in empirical scope for practical reasons. I assume thatgender is a unitary phenomenon across languages, and that its properties can beinvestigated in various languages with broadly generalizable conclusions. However,to be clear, this is an assumption. It is a separate and interesting research questionwhether some phenomenon that ostensibly fits the definition of gender may be in facta different phenomenon (see e.g. Wiltschko 2012 on reanalyzing Ojibwe gender asnominal aspect).

Another restriction in empirical scope is that the book focuses on languages withtwo or three genders. This was a conscious choice so that more space could bedevoted to topics like the interaction of gender and number, and the role of gender in


10 1 Introduction


nominalizations. However, the choice is not as limiting as it may initially seem. Two-and three-gender languages provide more than enough richness and complexity toget the analysis off the ground, push it forward, and ultimately test its predictions.Additionally, two- and three-gender languages are the most common types of gendersystems. Corbett (2011a) surveys 112 languages with gender systems, and of these, 75have two or three genders (67%). Nichols (1992) surveys 45 languages with gendersystems, of which 26 have two or three genders (57%). Since it can be reasonablyconcluded that two- and three-gender languages are the majority of languageswith gender systems, the analysis automatically covers most languages with gendersystems even when it is restricted to only these languages. However, I do not intendto downplay the potential contribution of languages with more than three genders,and I hope that the proposals here can be extended in future research tothese languages (see Chapter 11 for more specific thoughts on how this could beaccomplished).

Finally, this research is focused squarely within the theoretical linguistic traditionand will not engage directly with sociolinguistic or psycholinguistic approaches togender. However, in Chapter 11, I offer some final thoughts on how the results of thebook might be affected by social factors. Also, in the conclusion of Chapter 8, I lay outa promising connection between the results of this book and the conclusions of somepsycholinguistic work on gender.

1.4 The lay of the land: chapter previews

The book can be viewed as falling into three parts. The first part consists of twochapters on the gender system of the Ethiosemitic language Amharic; these chaptersserve to lay out the primary analysis of the book in detail. The Amharic gendersystem is mostly based on natural gender, but there are also a small number of nounswith arbitrary gender. Amharic’s gender system is hence different from most previ-ously analyzed gender systems, and it raises important questions about whethernatural gender features should be in the morphosyntax and if so, how they relateto arbitrary gender features. In Chapter 2, I describe its gender system and arguethat previous morphosyntactic approaches face serious difficulties in trying toaccount for it.

In Chapter 3, I introduce the n analysis in detail, as well as my assumption thatnatural gender features are interpretable and arbitrary gender features are uninter-pretable. I walk through the mechanics of it for Amharic, showing how it captures allthe generalizations identified in Chapter 2. I also present direct evidence fromAmharic that gender features are on n, looking at the interaction of gender andnumber, the interaction of gender and nominal vocalic pattern, and the role ofgender in nominalizations. The second half of Chapter 3 fleshes out the analysis,


1.4 The lay of the land 11


exploring the nature of licensing conditions and setting in context my assumptionsabout the interpretability of gender features.

The second part (Chapters 4–7) widens the empirical domain, showing how allthe two- and three-gender systems predicted by the analysis that I developed for Amharicare attested. Chapter 4 refines the definition of gender introduced in Section 1.1 as aprecursor for discussing gender in a cross-linguistic context. Chapter 5 focuses ontwo-gender languages with the simplest possible inventory of gendered ns given myassumptions. I provide two case studies of languages that have this inventory: Dieri(Pama-Nyungan; Australia), which has a masculine default gender, and Zayse andZargulla (Omotic; Ethiopia) which have a feminine default. I also identify a type ofgender system based on biological sex that is predicted never to occur, and I showhow one language that at first appears to have this gender system (Lealao Chinantec(Otomanguean; Mexico)) is instead based on animacy.

In Chapter 6, I focus on two-gender languages that contain at least one n withan uninterpretable (arbitrary) gender feature. I present three examples: Spanish(Indo-European (Romance); Spain, Central and South America), Maa (Nilo-Saharan(Nilotic); Kenya), and the Algonquian language family (Canada, United States).I argue that Spanish has a n with an uninterpretable feminine feature (like Amharic,which I compare it to), that Maa has a n with an uninterpretable masculinefeature, and that Algonquian languages have a n with an uninterpretable animacyfeature.

In Chapter 7, I turn to three-gender languages. I first examine Mangarayi (non-Pama-Nyungan; Australia), which has the simplest inventory of gendered ns. I chartthe differences and similarities between Mangarayi and the languages of Chapter 5 inthat Mangarayi has three genders whereas Dieri, Zayse, and Zargulla have twogenders. I then move on to Wari’ (Chapacuran; Brazil), which is the three-gendercounter part of Maa (Chapter 6) in that it has one uninterpretable masculine feature.Finally, I investigate Lavukaleve, a three-gender language that provides evidence fortwo ns with uninterpretable features. The chapter closes with an excursus on thecomplexities of default gender in three-gender languages.

The third part turns the focus to gender being on n, exploring the interaction ofgender and number (Chapter 8), the place of gender in nominalizations (Chapters 9and 10), and the relationship between gender and declension class (Chapter 10).

Chapter 8 centers on the fact that some nouns in certain languages change genderin the plural, i.e. a noun has Gender X in the singular and Gender Y in the plural.I argue that even though gender seems to be dependent on number in these cases,gender is not on the Num(ber) head in the syntactic derivation. I demonstrate howa gender-on-n proposal accounts for the fact that number seems to be conditioninggender via in-depth case studies of two languages: Somali (Afroasiatic (Cushitic);Somalia) and Romanian (Indo-European (Romance); Romania). For Somali, I arguethat all plurals are formed via n, and this explains why switching numbers can (but


12 1 Introduction


need not) involve switching genders. For Romanian, I argue in favor of the standardanalysis of the gender-switching nouns as unspecified for gender; I show how thisplays out in a n approach to gender, and how it results in the gender of such nounsbeing dependent on number.

In Chapter 9, I demonstrate how gender on n predicts that nominalizations will atleast be capable of being gendered. I provide data from a range of languages, bothones previously encountered in the book and ones not mentioned before, thatdemonstrates that this prediction is borne out. I further identify three correctpredictions of this approach, including that nominalizations that lack n shouldreceive the default gender. I then address two potential problems for the analysis,including the fact that gender features seem to be exponed separately from anominalizer in some languages, and I propose some solutions.

Chapter 10 has two goals, both related to further exploring the implications ofhaving gender features located specifically on n. First, I investigate nominals thatcontain multiple, stacked ns, i.e. multiple gender features. I show how the highest ndetermines the gender of the nominal, and develop an explanation of this fact basedon independently motivated assumptions about morphosyntactic cyclicity. I alsobriefly consider diminutive morphology and show how it supports the idea thatthe highest n determines the gender. The second goal of the chapter is to brieflyconsider declension class, which is often inserted post-syntactically at/near n in theDistributed Morphology literature. I demonstrate that declension class/pattern is notisomorphic with gender, but that gender can affect the choice of declension class—aspredicted if it is adjoined to n and gender is on n.

Chapter 11 concludes, returning to the questions about gender posed in Section 1.1and reviewing how the book has answered them. It also identifies key areas of futureresearch—open questions that will hopefully lead to further advances in our under-standing of the complex morphosyntax of gender.


1.4 The lay of the land 13


2

The Amharic gender system andprevious approaches to gender

2.1 Introduction

In order to introduce and justify a gender feature on n, I investigate over the nexttwo chapters the gender system of the language Amharic (Afroasiatic (Semitic);Ethiopia). In this chapter, I show how Amharic relies on both natural andarbitrary gender in its gender system (Section 2.2), in ways that are difficult forprevious gender approaches to explain (Section 2.3). In Chapter 3, I presentevidence from Amharic that n is the appropriate locus for gender, using datafrom the interactions of gender and number, gender and nominalization, andgender and declension class.

2.2 Gender in Amharic

I start with some basic background on Amharic as a language, since it is the focus ofthe next two chapters and Amharic data recur throughout the book. Genetically,Amharic is an Ethiosemitic language and a member of the Afroasiatic languagefamily. It is the national language of Ethiopia, taught in schools and used in nationalnewspapers and government publications.3 Lewis et al. (2013) report that there wereapproximately 21 million speakers of Amharic in Ethiopia according to the 2007census, with approximately 15 million monolingual. There are also significantAmharic-speaking diaspora populations in Europe, Israel, Canada, and the UnitedStates. All unattributed data are from my own fieldwork in the South San Francisco

The Morphosyntax of Gender. First Edition. Ruth Kramer.© Ruth Kramer 2015. Published 2015 by Oxford University Press.

3 There are approximately 80 languages spoken within Ethiopia, including 13 other Semitic languages(see above), many Cushitic languages (including Oromo, Sidamo and Afar), many Omotic languages,and several languages from the Nilo-Saharan family. In recognition of these many languages, Ethiopiahas no official language, but Amharic, with its “national” status, is by far the most prominent politicallyand is used as a lingua franca. For a wide-ranging discussion of language in Ethiopia, see Bender et al.(1976).



Bay region and the Washington, DC area, and the majority of the remaining data arefrom Leslau (1995), the foremost reference grammar of the language.4

Amharic has maintained a relatively low profile in theoretical generative linguis-tics, from the early studies of Bach (1970), Hetzron (1970), Fulass (1966, 1972), andManahlot (1977) to a cluster of work in the late 1980s and early 1990s (see e.g. Mullen1986, Yimam 1988, 1996 et seq., Tremblay and Kabbaj 1990, Halefom 1994, Amberber1996). In the past ten years, a formal semantic interest in Amharic has flowered due towork on indexical shifting (Schlenker 1999, 2003a, 2003b, Anand 2006), and acorresponding focus on Amharic syntax has taken shape (see e.g. Demeke 2001,2003, Ouhalla 2000, 2004, Henderson 2003, den Dikken 2007, Beermann and Ephrem2007, Yabe 2007, Eilam 2009, Kramer 2009, 2010, 2012, 2014a, 2014b, Kramer andEilam 2012, Baker 2012). None of these sources (with the exception of Kramer 2009,2014b) provides an analysis of the gender system, though, and I turn next to the factsof gender assignment.

2.2.1 The facts

Amharic distinguishes two genders, conventionally labelled “masculine” and “fem-inine” (Leslau 1995: 161). This is typical for a Semitic language (Rubin 2010: 34) andfor many, if not all, Afroasiatic languages as well (see e.g. Zaborski 1992, Hayward2000). There is no consistent morphophonological correlate of gender (Leslau 1995:161, Cohen 1970: 74), with one exception discussed below. Therefore, gender isindicated by agreement on e.g. definite markers, demonstratives, and verbs. Mascu-line and feminine definite markers are shown in (1).5

(1) Amharic definite markera. -u the.msb. -wa the.fs

When the gender of an Amharic noun needs to be indicated in this book, it willappear with the appropriate definite marker.

The Amharic system for assigning gender is heavily reliant on natural gender.Specifically, the gender of an animate noun is assigned exclusively according to itsnatural gender (Leslau 1995: 161ff., Hartmann 1980: 278ff., Appleyard 1995: 33). Somemale/female pairs have different roots; mostly these are kinship terms (father/mother) and domesticated animals (bull calf/heifer).

4 A handful of examples are from the Amharic internet (with the link an accompanying footnote), andfrom the Walta Information Center Tagged Amharic News Corpus (cited as from Walta; see Demeke andGetachew 2006 for more details on the corpus).

5 The definite markers could each be decomposed into a D morpheme /u/ and a gender marker. Themasculine definite marker would thus consist of /u+�/ and the feminine definite marker consists of /u+a/which surfaces as [wa]. In the analysis, I do not represent the definite marker as decomposed, in order tokeep the representations simple and in order to be in accordance with the previous Amharic literature.


2.2 Gender in Amharic 15


(2) Different-root nominals6

a. abbat ‘father’ h. �nnat ‘mother’b. bal ‘husband’ i. mist ‘wife’c. wänd�mm ‘brother’ j. �h�t ‘sister’d. aggot ‘uncle’ k. ak�st ‘aunt’e. wäyfän ‘bull calf ’ l. gidär ‘heifer’f. bäre ‘ox’ m. lam ‘cow’g. d�ngulla ‘stallion’ n. bazra ‘mare’

I am assuming that these pairs have different morphological roots, but similar pairsof items in other languages have sometimes been treated as suppletive (see e.g.Osthoff 1899, Markey 1985, Mel’čuk 1976). However, there is evidence that thesepairs are not suppletive from the relevant non-gendered term, e.g. ‘parent,’ ‘calf,’ etc.The non-gendered term in almost every case above is (i) either a phonologicallydistinct form (e.g. t’i�dʒdʒa ‘calf ’, wäladʒ ‘parent’) or (ii) unattested (e.g. there is noequivalent of the English word sibling). Either scenario is suspicious from theperspective of a suppletive analysis; the purported root shared by the pair wouldeither leave no morphophonologic trace across three slots of the paradigm or itwould never surface with a non-gendered meaning. I conclude that the pairs in (2)are not morphologically related via suppletion, and continue to refer to them asdifferent-root nominals.

Different-root nominals are only a small subset of the animate nominals inAmharic.7 The vast majority have the same root for both males and females, whetherfor humans (e.g. tämari ‘student’) or animals (e.g. wi�ʃʃa ‘dog’). The female version isfeminine and the male version is masculine.

(3) Same-root nominalsa. tämari-w8 tämari-wa

student-def.m student-def.f‘the (male) student’ ‘the (female) student’

6 I do not classify säw ‘man’ and set ‘woman’ as different-root nominals. They are derived etymologic-ally from the same Ge’ez root /sb?/ (unlike most of the different-root nominals, e.g. wändi�mm and i�hi�tcome from different roots; see Leslau 1969). They also behave differently from other different-rootnominals in that the male form (säw) can be used for the superordinate term, i.e. with the meaning‘person’ (it is common cross-linguistically for ‘man’ also to be used as a superordinate for ‘person’; I assumethat the choice of the male form as the default is due to social factors: see Ch. 11). Neither the male form northe female form can be used as the superordinate for the other different-root nominals, even when there isnot a lexicalized form of a gender-neutral term (e.g. when there is no word for ‘sibling’).

7 There are also a few nominals that behave like different-root nominals but are not paired, i.e. they arelimited to one gender. For example, mäsfi�n ‘duke’ is only used for males and there is no counterpart‘duchess.’

8 The masculine definite marker -u is realized as -w after a vowel.


16 2 The Amharic gender system

hodgecInserted Textal

RSticky NoteYes, this should be "morphophonological"


b. mu ʃ�rra-w mu ʃ�rra-wawedding.participant-def.m wedding.participant-def.f‘the groom’ ‘the bride’

c. hakim-u hakim-wadoctor-def.m doctor-def.f‘the (male) doctor’ ‘the (female) doctor’

d. halafi-w halafi-waperson.in.charge-def.m person.in.charge-def.f‘the (male) person in charge’ ‘the (female) person in charge’Walta hed12a2 Walta hed01a29

e. w� ʃ ʃa-w w� ʃ ʃa-wadog-def.m dog-def.f‘the (male) dog’ ‘the (female) dog’

f. awraris-u awraris-warhinoceros-def.m rhinoceros-def.f‘the (male) rhinoceros’ ‘the (female) rhinoceros’

Several kinship terms are also same-root nominals including ayat ‘grandparent,’amat ‘parent-in-law,’ warsa ‘sibling-in-law,’ and zämäd ‘relative.’

The default gender is masculine. For example, if the natural gender of the referentis unknown, then the nominal is masculine. In (12), the speaker does not know thenatural gender of the baby, but uses a masculine definite marker.

(4) h�s’an-u wänd näw set?baby-def.m male be.3ms.s female?‘Is the baby a he or a she?’10 (Leslau 1995: 164)

Additional evidence for a masculine default is that the nominal ‘nobody’ takesmasculine agreement (cf. Roca 1989).

(5) balläfäw samm�nt betäkr�stiyan mann�mm al-hed-ä-mmlast week church nobody neg-go-3ms.s-neg‘Last week, nobody went to church.’ (Leslau 1995: 122)

Exceptionally, certain animals are feminine if their gender is unknown/irrelevant(Leslau 1995: 166, Hartmann 1980: 281, Cohen 1970: 75).11

9 These examples are from theWalta Information Center Tagged Amharic News Corpus (Demeke andGetachew 2006). See fn. 4.

10 The noun hi�s’an ‘baby’ is a same-root nominal, i.e. it can be either masculine or feminine dependingon whether it refers to a male or female infant.

11 Other such animals include i�bab ‘snake,’ asa ‘fish,’ ni�b ‘bee’ (all from Leslau 1995) as well as zi�nb ‘fly’and t’i�ntʃäl ‘rabbit’ (from my own fieldwork).




(6) Feminine-default nounsa. bäk’lo-wa b. ayt’-wa c. k’äbäro-wa d. ʃärärit-wa

mule-def.f mouse-def.f jackal-def.f spider-def.f‘the mule’ ‘the mouse’ ‘the jackal’ ‘the spider’

If the natural gender of the referent for one of these animal nouns is known, though,it overrides the feminine default.

(7) ayt’-umouse-def.m‘the male mouse’

This demonstrates that natural gender, if known, always determines the gender of ananimate nominal.

As for inanimate nominals, almost all of them are masculine (Leslau 1995: 161,Cohen 1970: 74).

(8) Masculine nouns (inanimate)a. mot ‘death’ f. w�d�dd�r ‘competition’b. k�b�r ‘honor’ g. bet ‘house’c. wänbär ‘chair’ h. d�mm�r ‘total, sum’d. d�ngay ‘stone’ i. wäräda ‘district’e. k�bäb ‘circle’ j. gazet’a ‘newspaper’

Only a handful of inanimate nouns are treated as feminine; some examples are givenin (9).

(9) Feminine nouns (inanimate)a. mäkina ‘car’ e. s’ähay ‘sun’b. azurit ‘whirlpool’ f. kätäma ‘city’c. agär ‘country’ g. betä kr�stiyan ‘church’d. m�d�r ‘earth’ h. tʃ ’äräk’a ‘moon’

It is difficult to calculate the exact number of feminine nouns, since nouns are notlisted in Amharic dictionaries with their gender (which indicates how small a rolearbitrary gender plays in Amharic). After surveying the gender sections of threegrammars (Leslau 1995, Hartmann 1980, Cohen 1970), as well as performing somebasic searches in the Walta Information Center Tagged Amharic News Corpus(Demeke and Getachew 2006), my best estimate is that there are about twenty tothirty feminine nouns.

As for gender morphology, masculine gender is never morphologically marked(unsurprisingly). Feminine gender is also not universally associated with a particularaffix, unlike in some other Afroasiatic languages like Ancient Egyptian (where femininegender is marked by a -t suffix; see e.g. Gardiner 1957). A feminine -t or -at suffix has




been reconstructed in Proto-Afroasiatic and is one of the common features acrossAfroasiatic languages in general (see e.g. Zaborski 1992, Hayward 2000).A descendant of this suffix remains in Amharic: in some same-root animate nom-inals, the female form can take the suffix -it (Leslau 1995: 163–4, Hartmann 1980: 280).

(10) Same-root nominals with -ita. l�dʒ l�dʒ-it e. mämh�r mämh�r-t

‘boy, child’ ‘girl’ ‘teacher’ ‘female teacher’

b. mänäkwse mänäkws-it12 f. t’ot’a t’ot’-it‘monk’ ‘nun’ ‘ape’ ‘female ape’

c. ʃ�mag�lle ʃ�mag�ll-it g. �gäle �gäl-it‘old man’ ‘old woman’ ‘so-and-so’ ‘female so-and-so’

d. mu ʃ�rra mu ʃ�rr-it‘groom’ ‘bride’

This is not a highly productive process, since it is not possible for all animate nouns,e.g. *tämar-it ‘female student’ and *hakim-it ‘female doctor.’ It is also not determin-istic; a noun can be feminine without -it even if it could have the -it suffix: (3)bmuʃi�rra-wa ‘(the) bride’ is just as good as (10)d muʃi�rrit.

In general, the suffix -it is neither a necessary nor a sufficient condition forfeminine gender in Amharic, regardless of animacy (Leslau 1995: 163–4, Cohen1970: 74). Nouns can be feminine without ending in -it, and nouns can be masculineand end in -it.

(11) Feminine, no -it Masculine, end in -its’ähay ‘sun’ kulalit ‘kidney’agär ‘country’ särawit ‘army’tämari-wa ‘the (female) student’ mogzit-u ‘the (male) tutor’ (Cohen 1970: 74)

Finally, the suffix -it does not convert inanimate nominals to feminine arbitrarygender. In Hebrew, adding a feminine suffix (-et, -it) to an inanimate masculine nounderives a semantically related feminine noun (Ritter 1993).

(12) a. magav magev-et Hebrew‘wiper’ ‘towel’

b. maxsan maxsan-it‘warehouse’ ‘magazine’ (Ritter 1993: 796, (2))

12 The final vowels in these nouns are deleted when the -it suffix is added in order to avoid hiatus. Thisis similar to other kinds of nominal suffixes, which also trigger deletion of the final vowel on the stem whichthey attach to (Leslau 1995: 36).




For example, magav without any suffixes has the meaning ‘wiper,’ but magev with afeminine suffix -et means ‘towel.’ In Amharic, however, adding -it to an inanimatemasculine noun results in a diminutive interpretation of the nominal13 (i.e. adding aninterpretation that the nominal is small and/or cute, among other readings; seeLeslau 1995: 167–8), not a new, semantically related nominal.

Moreover, if the “feminine suffix” is removed from nominals that end in -it likeʃärärit ‘spider’ or azurit ‘whirlpool,’ the result is not a related masculine noun. Thereis in fact no such word as ʃärär in Amharic, and azur is a verbal form (the masculineimperative of the verb zorä ‘to turn,’ which is morphologically related to azurit). Insum, then, the suffix -it has a rather limited role: it is a non-productive means ofindicating only female natural gender only on certain same root nominals.

A last piece of gender-related morphology is the set of gender “specifiers” thatindicate natural gender (Leslau 1995: 164–6, Cohen 1970: 76, Hartmann 1980: 279). Forhuman nouns, the specifiers are wand for males and set for females.

(13) a. wänd ayat male grandparent ‘grandfather’b. set ayat female grandparent ‘grandmother’

There are a few additional specifiers only for animal nouns which denote either maleor female gender, but the lexical items are different from those used for humannouns. There seem to be two options for analyzing the gender specifiers. First, theycould be adjectives, like ‘male’ and ‘female,’ but more differentiated than in, say,English. Second, they could be nominal classifiers, similar to those found in Mayanlanguages, Bantu languages and many other language families (see Chapter 4).

There is some indication that the adjective analysis is correct. The specifiers exhibitthe same morphosyntactic behavior as adjectives (e.g. the definite marker attaches tothem; Leslau 1995: 165) and they can be predicates of a copular clause (see (4)). Also,most classifier systems operate over several more criteria than gender (animacy,shape, size, etc.) and do not usually coexist with a masculine/feminine two-gendersystem. I thus assume the gender specifiers are adjectives, and do not treat themfurther.

2.2.2 Summary, typology, diachrony

Looking at the facts as a whole, Amharic assigns gender mostly based on naturalgender, i.e. biological sex (or the lack thereof). Arbitrary gender is only relevant inAmharic for the small number of feminine inanimate nouns, and the even smallernumber of feminine-default animals. Otherwise, the natural gender (or lack thereof)

13 This is true under the assumption that, in a nominal like bet-it-u ‘the small house’, -it is the feminine/diminutive suffix and not part of the definite article. In Leslau (1995), it is claimed that the feminine definitemarker may surface as -itu (also -itwa), but it is unclear whether these forms are truly feminine definitemarkers or combinations of -it and a definite article.




determines the gender of a nominal in Amharic according to the descriptive rules setout in (14):

(14) Gender assignment rules in Amharica. If a nominal refers to a male referent, the nominal is masculine.b. If a nominal refers to a female referent, the nominal is feminine.c. If a nominal refers to a referent whose natural gender is unknown, or which

does not have natural gender, the nominal is masculine by default.

It is useful to consider briefly how Amharic fits into Corbett’s (1991) classification ofthe systems of gender assignment in the world’s languages. Corbett draws a funda-mental distinction between semantic systems of assignment, where most nouns areassigned gender according to semantic principles, and formal systems of assignment,where most nouns are assigned gender according to morphological or phonologicalprinciples. Both kinds of systems are found in a variety of languages and languagefamilies, and semantic and formal criteria can overlap in a particular language (seeChapter 11 for further discussion).

Amharic is best described as either a “strict semantic” or a “predominantlysemantic” system (Corbett 1991: 13), where the gender of most nouns is assignedvia semantic principles but there are certain sets of exceptions.14 It is certainly not thecase that phonology or morphology determine the gender of a noun in Amharic—there are no phonological regularities about which nouns are assigned which genderand the only morphological indication of gender (the -it suffix) is neither necessarynor sufficient to deduce gender.

The semantic principles are clear, though, as stated in (14). (14)ab in particularseem to be virtually exceptionless.15 There are, of course, a small number of excep-tions to (14)c—there is a “residue” of inanimate nouns that are assigned femininegender. Similarly, there are a small number of animal nouns for whom feminine isused when the gender is unknown and which (again) must be memorized.

The small residue of inanimate feminine nouns suggests that Amharic may havehad a gender system based on arbitrary gender in the past, and changed to one morebased on natural gender over time. In this respect, it is illustrative to look at thehistory of gender in English. In the past millennium, English changed from alanguage that relied on arbitrary gender and natural gender to a language that reliesalmost exclusively on natural gender (i.e. in Modern English, inanimate nominals arenot sorted into two or more arbitrary genders, and pronouns and nominals referring

14 It is unclear whether the number of exceptional nouns in Amharic is enough to render it “predom-inantly semantic.”

15 The major exception is the use of diminutive forms (which are all feminine) to refer to male animates(with some kind of emotional impact: affection, mockery, etc.). See Chs 3, 9, and 10 for more informationon diminutives.




to animate nominals have the natural gender of their referents; for recent perspec-tives on the shift, see Curzan 2003 and Platzer 2005). It is commonly believed that theloss of gender morphology on both nouns and modifiers caused or at least greatlyabetted the loss of arbitrary gender. Without morphological cues about nominalgender, it is difficult for a learner to determine (i.e. acquire) the gender of a nominal.

The ancient Ethiosemitic language Ge’ez (spoken during the Axumite empire, firstwritten down around the 4th century bce, now the liturgical language of theEthiopian Christian church) is not a direct ancestor of Amharic; it belongs to theNorth branch of Ethiosemitic with Tigre and Tigrinya, whereas Amharic is part ofthe South branch with Harari and the Gurage languages. However, it offers sometantalizing clues about what an earlier stage of Amharic might have been like, andsuggests that changes occurred which facilitated the loss of arbitrary gender inAmharic.16

Ge’ez had both masculine and feminine gender, but it is difficult to ascertainwhether there were more feminine inanimate nouns than there are in Amharic.17 It isclear though, that Ge’ez had pervasive gender agreement. Adjectives generally agreedin gender (Lambdin 1978: 68), many (if not all) of the verbs show distinct feminineand masculine forms in the singular and in the plural (see e.g. the perfect verbalparadigm in Lambdin 1978: 50, and the imperfect verbal paradigm in Lambdin 1978:144), and cardinal numbers showed gender agreement. Amharic, in contrast, does nothave consistent gender agreement for adjectives, and has lost gender agreemententirely on plural verbs and cardinal numbers. There is still some gender agreementin Amharic (on definite markers, demonstratives, and singular verbs), but the loss ofsome gender agreement might have facilitated or triggered a shift away from arbi-trary gender, similar to the loss of agreement in the history of English.

Moreover, there is some evidence for this scenario in that the shift to naturalgender is nearly complete for younger speakers (less than 25 years old). They tend totreat any feminine inanimate noun as a diminutive (see Chapter 3), and they treatfeminine-default animal nominals like same-root nominals (i.e. with a masculinedefault). This means some younger speakers have not acquired arbitrary gender; they

16 There are written records of earlier forms of Amharic, some dating as far back as the 1300s. However,Demeke (2013) does not note any differences between the gender system of Old Amharic (pre-18th c.) andModern Amharic. Therefore, it is necessary to look further back in time to speculate about the origin of theAmharic gender system.

17 Lambdin (1978: 26–7) and Tropper (2002: 69–70) observe that the gender of certain inanimate nounsseems to vary in Ge’ez; but this statement should be taken with a grain of salt. Gender usage varies acrosstexts (meaning that gender may be consistent for a particular author) and some feminine nominals appearto be expressive in use (i.e. diminutives). Both these factors could seriously inflate the number of wordswhose gender seems to vary. Tropper also observes that one gender tends to be dominant for any givenword, increasing the chance that variation may be due to expressiveness or individual authoridiosyncrasies.




treat any non-female-denoting noun as masculine by default, and thus the Amharicgender system for them is entirely semantic.

In sum, the Amharic gender system heavily relies on natural gender (or the lackthereof ) to assign genders, and possibly evolved from a system with more arbitrarygender in Ge’ez via the loss of gender agreement. In the next section, I start to tacklehow to morphosyntactically analyze gender in Amharic—reviewing several previousanalyses to show how they struggle with the Amharic facts.

2.3 The morphosyntax of gender: previous approaches

In mainstream syntactic theory, agreement is treated as a relation which is estab-lished during the syntactic portion of the derivation. Thus, in order for genderagreement to occur, gender features must be present during the narrow syntax (seealso Pfau 2009 for empirical evidence from speech errors that gender features are inthe syntax). If gender features must be present in the syntax, though, a gender featuremust be one of the features on/of some terminal node in the syntactic derivation. Thequestion is: which one? This is one of the driving questions of the book: what is thelocus of gender features in the syntax?18

Intuitively, gender seems to be an inherent property of nouns. The gender of anoun is generally consistent no matter how it is inflected: for case, number, defin-iteness, etc.19 This intuition has led to many proposals that the gender feature is inthe lexical entry for any given noun, i.e. on the nominal head N in the syntax, asshown schematically in (15) for a feminine noun.

(15) DP

D NP

N [+fem]

However, it has occasionally been claimed that the gender feature can or mustoriginate elsewhere than on the noun, and I discuss these proposals first inSection 2.3.1. I then return to the idea that gender is on N, first from a lexicalistviewpoint (Section 2.3.2) and then from a Distributed Morphology perspective(Section 2.3.3).

18 Under a post-syntactic approach to agreement (e.g. Bobaljik 2008), gender features could be insertedin the morphology before the agreement relation as dissociated features. The question would remain,t

faculty.georgetown.edufaculty.georgetown.edu/rtk8/9780199679935-kramer-revises1.pdf · comp. by:...

Documents